The effect of high- and low-frequency previews and sentential fit on word skipping
during reading
Bernhard Angele1, Abby Laishley1, Keith Rayner2, and Simon P. Liversedge3
1
Psychology Research Centre, Faculty of Science and Technology, Bournemouth University
2
Department of Psychology, University of California San Diego
3
School of Psychology, University of Southampton
Correspondence to:
Bernhard Angele
Psychology Research Centre
Faculty of Science and Technology, Bournemouth University
Talbot Campus
Fern Barrow, Poole BH12 5BB, United Kingdom
Email: [email protected]
Running Head: Skipping of short words
2
Abstract
In a previous gaze-contingent boundary experiment, Angele and Rayner (2012) found
that readers are likely to skip a word that appears to be the definite article the even when
syntactic constraints do not allow for articles to occur in that position. In the present study, we
investigated whether the word frequency of the preview of a three-letter target word influences a
reader’s decision to fixate or skip that word. We found that the word frequency rather than the
felicitousness (syntactic fit) of the preview affected how often the upcoming word was skipped.
These results indicate that visual information about the upcoming word trumps information from
the sentence context when it comes to making a skipping decision. Skipping parafoveal instances
of the therefore may simply be an extreme case of skipping high-frequency words.
3
In order to allocate their gaze efficiently, readers have to anticipate how much
information they will need about upcoming (that is, not yet fixated) words. There are two
possible sources for information about upcoming words: First, readers can use parafoveal vision
to preprocess words before they are fixated (McConkie & Rayner, 1975; Rayner, 1975).
However, there is debate about the extent and the quality of the parafoveal information that is
available when the oculomotor system makes a decision about where to fixate next and when to
make the next saccade (see Liversedge & Findlay, 2000; Rayner, 1998, 2009; Rayner &
Liversedge, 2011; Schotter, Angele, & Rayner, 2012 for reviews)1. Second, readers can also use
the preceding sentence context to predict/guess (at an unconscious level) the identity of the
upcoming words (Balota, Pollatsek, & Rayner, 1985; Bicknell & Levy, 2010; Ehrlich & Rayner,
1981; Levy, Bicknell, Slattery, & Rayner, 2009; Rayner & Well, 1996). However, such
predictions/guesses may not always be correct and may require input from high-level sentence
integration processes which might not be available early enough to influence oculomotor
decisions. The question we will consider is to what extent readers use parafoveal information and
sentence context information when they make a skipping decision.
In determining the relative importance of parafoveal information compared to
predictability from the sentence context, the phenomenon of word skipping is particularly
interesting: if a word is intentionally skipped, this can be taken to imply that it has been
parafoveally processed to the point where it no longer needs to be fixated. This assumption is
explicitly implemented in serial attention shift models of eye-movement control during reading
such as the E-Z Reader model (Reichle, Pollatsek, Fisher, & Rayner, 1998; Reichle, Warren, &
McConnell, 2009), according to which a word is never intentionally skipped unless it has been
parafoveally identified. This view derives from the assumption that words are processed one at a
4
time and in the correct order (but not necessarily fixated in that order). Word skipping in E-Z
Reader occurs because readers start preprocessing the upcoming word parafoveally whenever
they have identified the currently fixated word but their oculomotor system is not yet ready to
perform a saccade to the next word. If the upcoming word is very easy to process, it is possible
that it too is identified before the saccadic program has completed. In this case, the current
saccadic program is cancelled and a saccade to the second word to the right—that is, a skipping
saccade—is prepared2.
A different approach to word skipping – also stressing the importance of parafoveal
information – is made by processing gradient models such as the SWIFT model of eyemovement control (Engbert, Nuthmann, Richter, & Kliegl, 2005), which assume that multiple
words can be processed in parallel and therefore allows for word skipping even if a word has not
been fully identified. Specifically, SWIFT proposes a dynamic field of activations with one
activation value for each word in a sentence. The maximum activation value of each word is
determined by its word frequency. Initially, the further a word has been processed, the higher its
activation value is (predictable words rise in activation more slowly than unpredictable words).
Once the activation of a word reaches its maximum, further processing lowers the activation
value until it is back at zero at which point the word is assumed to be fully identified. In SWIFT,
saccades are triggered at random intervals (with a possible delay due to foveal word difficulty).
The probability of a word becoming the target of the next saccade is directly proportional to its
activation level relative to the activation levels of the other words in the field. This means that
any word that has been processed to any degree and that is not yet fully identified can be a
potential saccade target in SWIFT. Because of this, skipping the upcoming word is possible as
soon as the word after it has accumulated some activation. For example, if the upcoming word
5
has a relative activation of .7 and the word to its right has a relative activation of .2, SWIFT will
skip the upcoming word in 20% of simulations (and fixate it in 70% of simulations).
Importantly, both E-Z Reader and SWIFT allow for an influence of word predictability
from the sentence context on skipping probability. However, neither model is very clear on how
the oculomotor system estimates predictability. In particular, both models assume that, in order
for predictability to play a role in word identification, the word in question must have received at
least some parafoveal preprocessing – simply guessing a word without any parafoveal input does
not occur according to either model. Finally, both the SWIFT and E-Z Reader models can also
account for accidental word skipping due to oculomotor error.
Much of the research on word skipping has focused on those words that are skipped most
often during reading, such as the definite article the in English or the plural definite article les in
French. O’Regan (1979) and Gautier, Le Gargasson, and O’Regan (2000) found that les was
skipped more often than other three-letter words even when it was not predictable from the prior
context. Angele and Rayner (2012) investigated this “automatic” the-skipping phenomenon
further by using the gaze-contingent boundary paradigm (Rayner, 1975) to present readers with
infelicitous (syntactically illegal) previews of the article the. This was done by having subjects
read sentences containing three-letter target verbs (e.g. ate). In the critical condition, while
fixating to the left of the target word, the parafoveal preview for the target verb was replaced
with the article the. This ensured that all the-previews were infelicitous in the sentence context.
There were two additional conditions, one in which the preview was correct (control condition)
and a random letter nonword preview condition.
6
After crossing the boundary, readers always saw the correct word. Angele and Rayner
found that readers skipped nearly 50% of the words that appeared as the, even when the sentence
context did not syntactically allow an article in the target word position. This skipping rate was
comparable with that calculated for felicitous occurrences of the in other positions in the
sentences, suggesting that, in accordance with Gautier et al.’s results, skipping of articles is not
strongly influenced by constraints of the preceding sentence context. In other words, Angele and
Rayner’s results seem to indicate that word skipping is mainly influenced by parafoveal
processing. However, it is not clear whether article previews are skipped more often than correct
verb previews because of their higher word frequency alone, or whether the lower fixation
probability is specific to the article the (and perhaps a small number of other common, short
function words).
In order to investigate this important theoretical question further, we used a paradigm that
was quite similar to the one used by Angele and Rayner (2012). Specifically, we used three-letter
target words (dog in the sentence “The excitable dog was ready to go for his walk”), the
previews of which could either be correct and compatible with the sentence context (identical
control condition, dog), a random letter string (fze), or a higher- or lower-frequency three-letter
word that did not fit in the sentence context (dim). According to the CELEX corpus (accessed
using the N-Watch software by Davis (2005), the low frequency targets had a mean word
frequency of 10 (SD = 14), while the high frequency words had a mean word frequency of 1177
(SD = 3110). The mean difference in frequency within each pair was 1168 (SD = 3108; see Table
A2 in the Appendix for details). It is important to note that, in the above example, presenting an
adjective preview instead of the target word (which was a noun) caused a syntactic violation.
This was the case in the majority of our stimuli (see the Method section for details). Due to their
7
limited horizontal extent, such short words should be skipped quite often in reading (Hautala,
Hyönä, & Aro, 2011; Rayner & McConkie, 1976). On the other hand, high frequency words are
skipped more often than low frequency words (White, 2008). As a consequence, we expected
there to be a number of potential influences on fixation probability and fixation times.
First, the sentence context is very likely to have a strong influence on all eye-movement
measures. However, it is important to note that the main effect of sentence frame is not readily
interpretable as the pre- and post-target words necessarily differed between sentence frames.
Even on the target word, the effect of the context might be influenced by spillover effects from
the previous words. Consequently, in addition to the main analysis across sentence frames, we
will present separate analyses for those sentence frames in which the target was a high-frequency
word (and the dissimilar previews could either be a random letter string or a lower frequency
word) and those sentence frames in which the target was a low-frequency word (and the
dissimilar previews could be a random letter string or a higher frequency word).Within sentence
frames, the context was constant across all preview conditions.
Second, the parafoveal preview should have a clear effect on the processing of the target
word. Strong preview benefit effects on the target word were predicted, that is, shorter fixation
times when the target word preview was identical to the target word and longer fixation times
when it was dissimilar. We also anticipated that these effects might spill over onto the post-target
word.
Finally, we expected the parafoveal preview to have an immediate effect on eye
movement behavior before fixating the target word: If the preview is a non-word, subjects should
be more likely to fixate the target word. The presence of a non-word in the parafovea may also
8
lead to longer fixation times on the pre-target word (such an effect of parafoveal orthographic
information on ongoing processing as evidenced by the duration of the current fixation is known
as an orthographic parafoveal-on-foveal effect; see Schotter et al., 2012). For those conditions in
which the preview is a word (that is, the identical and alternative preview conditions), the
probability of skipping the target word could be affected by the frequency of the preview and by
the fit between preview and sentence context. If readers only take the frequency of the upcoming
word into account, the frequency preview should have a direct influence on whether the target
word is skipped or fixated independent of the sentence context, with high-frequency previews
leading to higher skipping rates than low-frequency previews. Alternatively, if readers take the
context into account when making a skipping decision, they should be less likely to skip a
syntactically invalid preview than a syntactically valid preview. Of course, target word skipping
might be affected by both preview word frequency and contextual fit at the same time.
Additionally, based on previous studies (Kennedy, 1998; Kennedy, Murray, & Boissiere, 2004;
Kennedy & Pynte, 2005; Kennedy, Pynte, & Ducrot, 2002; Kliegl, Nuthmann, & Engbert, 2006),
a low-frequency word in the parafovea might also lead to longer fixation times on the pre-target
word than a high-frequency word (a lexical parafoveal-on-foveal effect). In more general
theoretical terms, the outcome of this experiment will tell us more about the importance of
predictive linguistic processing during reading.
Method
Subjects
Thirty-five students from the University of Southampton participated in the experiment
for course credit. All were native English speakers with normal/corrected-to-normal vision and
no diagnosed reading difficulties.
9
Apparatus
Eye-movements were monitored every millisecond using an SR Research Eyelink 1000
eye-tracker. Viewing was binocular, though eye movement data were only collected from the
right eye. Sentence stimuli were displayed on a computer monitor with a refresh rate of 150 Hz3.
Viewing distance was 73 cm, with 3 characters equaling 1° of visual angle. A video-game
controller was used by subjects to end each trial and respond to comprehension questions.
Materials
We assembled a list of 120 pairs of three-letter high- and low-frequency words; members
of each pair differed in word frequency. Furthermore, we constructed 240 sentence frames (see
Appendix for details). Each sentence frame was compatible with either the high or the low
frequency word of a pair, but not with the other word of that pair. For most of the word pairs and
sentence frames, this incompatibility was syntactic (144 out of 240), but, due to the difficulty
inherent in finding word pair/sentence frame combinations resulting in true syntactic anomalies,
about one third of the word pair/sentence frame combinations instead had semantic
incompatibilities (It was an emotional day/dew for all of the family members, 96 out of 240).
Some of the target word pairs also contained content words (e.g. can/cow). We will present a
number of supplementary analyses aimed at determining whether these differences between
items had an impact on the effect of the preview manipulation at the end of the next section.
Each subject read only one of the two sentence frame versions, resulting in an equal number of
high-to-low and low-to-high frequency manipulations (see Figure 1 for an example).
Insert Figure 1 about here
10
To summarize, there were three preview conditions: the preview word was either
identical to the target word (dog/dog; dim/dim), a dissimilar random letter-string (fze/dog;
fzj/dim), or the infelicitous lower/higher frequency alternative word (dim/dog; dog/dim). Each
subject read a total of 120 sentences.
Procedure
Subjects were asked to read each sentence for comprehension. A chin/forehead rest was
used to minimize participant head movements. Initial three-point calibrations were carried out
until error was <0.3° and re-calibrations were completed as needed. The entire session lasted
approximately 45 minutes4.
Results and Discussion
We computed a number of standard eye movement measures (Rayner, 1998, 2009) on the
pre-target word (e.g. excitable), the target word (dog), and the post-target word (was).
Specifically, we computed two early processing time measures, first fixation duration (FFD;
mean duration of the first fixation on a word) and gaze duration (GD; the sum of all fixations on
a word before leaving it), both calculated only for words that were not initially skipped. Most
importantly, we computed the probability of skipping the target word as well as the probability of
making a regression out of the target word. The latter measure only showed significant effects on
the post-target word. We therefore do not report the results from the regression probability
analyses in the target and the pre-target word regions. Finally, we also calculated two later
processing measures, go-past time (Go-past; the sum of all the fixations on a word and any
regressive fixations before moving to the right of the target word) and total time (TT; the sum of
all fixations on a word during a trial). Tables 1a through 3a show the means and standard
deviations for each condition and each of the dependent variables.
11
In the present study, 28.6 % of the display changes completed more than 10 ms after the
onset of the subsequent fixation. In this case, we discarded the corresponding trial from the
analysis. This was done because previous research (Slattery, Angele, & Rayner, 2011) indicated
that display changes delayed by more than 10 ms cause a change in eye-movement behavior
even when subjects are not aware of them. Furthermore, if a fixation was shorter than 80 ms and
located within one character space (11 pixels) of another fixation, it was merged into that
fixation, otherwise it was deleted. Fixations with durations that deviated from a subject’s mean
by more than two standard deviations were deleted as well (around 5% of the data). All subjects
answered at least 85% of the comprehension questions correctly.
We expected that the frequent skipping of the three-letter target words and exclusion of
delayed display changes would lead to unequal cell sizes, which would make the use of
ANOVAs to analyze the data difficult. Thus, we used linear mixed models (LMM) with the
target word preview condition (identical vs. alternative vs. dissimilar) and the sentence frame
used (high frequency target word vs. low frequency target word) as well as their interaction as
fixed effects; additionally, the model contained random intercepts and random slopes for preview
condition and sentence frame condition (Baayen, Davidson, & Bates, 2008)5. For the preview
condition, we used successive differences contrasts, comparing the identical with the alternative
and the alternative with the dissimilar condition. Since the sentence frames differed in many
ways, interpreting the main effect of sentence frame is not possible. However, the interaction of
sentence frame with preview is still interesting, especially on the target word. In order to further
investigate cases where this interaction reached significance, we performed separate analyses for
each of the two sets of sentence frames. We used the lmer function from the lme4 package
(Bates, Maechler, Bolker, & Walker, 2013) within the R Environment for Statistical Computing
12
(R Development Core Team, 2013) to fit the LMMs. For each factor (preview word, sentence
frame, and their interaction), we report regression coefficients (b), standard errors, and t-values.
For binomial dependent variables such as fixation and regression probabilities, we report
regression coefficients, standard errors, and z-values from generalized LMMs using a logit-link.
It is not clear how to determine the degrees of freedom for the t-statistics estimated by the
LMMs, making it difficult to estimate p-values (Baayen et al., 2008). However, since our
analyses contain a large number of subjects and items and only a few fixed and random effects
are estimated, we can assume that the distribution of the t-values estimated by the LMMs
approximates the normal distribution. We therefore used the two-tailed criterion |t| ≥ 1.96 which
corresponds to a significance test at the 5% α-level. The z-values from the generalized LMMs
can be interpreted in exactly the same way. Tables 1b through 3b (sentence frames with highfrequency target word) and 1c through 3c (sentence frames with low-frequency target word)
show the LMM results, although the coefficient estimates and statistics for significant effects are
also repeated in the text.
Insert Tables 1-3 around here
Early processing measures: Skipping probability
Pre-target word. The probability of skipping the pre-target word was influenced by the
preview condition; readers were significantly less likely to skip the pre-target word in the
alternative condition than in the identical condition (b = .45, SE = .18, z = 2.49). Readers were
also less likely to skip the pre-target word in the dissimilar preview condition than in the
alternative preview condition (b = -.44, SE = .19, z = -2.3). These effects, which are similar to
effects observed by Angele and Rayner (2012), were unexpected and it is not quite clear what
causes them. Perhaps readers are on some occasions able to detect an anomaly arising from the
13
preview manipulation on the target word while fixating two words to the left of it (see Rayner,
Angele, Schotter, & Bicknell (2013) for discussion), though a fair amount of prior research
indicates this is unlikely (see Rayner, 1998, 2009 for reviews). Consistent with our last point,
note, however, that skipping of this region only occurred on a minority of trials (between 11 and
17%)—in the majority of trials, readers either do not notice the parafoveal violation or do not
react to it until they reach the pre-target word.
Target word. Both the difference in skipping probability between the identical and
alternative conditions (b = .52, SE = .21, z = 2.5) and the difference between the alternative and
dissimilar conditions (b = -.61, SE = .21, z = -2.91) were modulated by sentence frame.
Specifically, sentence frames with high-frequency targets showed a small, and, likely due to lack
of power, only marginally significant difference between the alternative and the identical
condition, with the target word being less likely to be skipped in the alternative than in the
identical condition (b = -.29, SE = .17, z = -1.69), and barely any difference at all between the
dissimilar and alternative conditions (b = .00, SE = .18, z = -.02). In contrast, sentence frames
with low-frequency targets showed a small difference in the opposite direction between the
alternative and the identical conditions, with the target being skipped less often in the identical
than in the dissimilar condition (b = .25, SE = .15, z = -1.64), and a very strong difference
between the dissimilar and the alternative conditions, with the dissimilar condition resulting in
many more target fixations than the alternative condition (b = -.62, SE = .15, z = -4.17). This
interaction suggests that, overall, readers were more likely to skip a target word with a highfrequency preview (the identical preview for high-frequency target sentence frames and the
alternative preview for low-frequency target sentence frames) than a target word with a low-
14
frequency preview (the alternative preview for high-frequency target sentence frames and the
identical preview for low-frequency target sentence frames), confirming our prediction.
Post-target word. There was a significant interaction between sentence frame and
preview on skipping probability on the post-target word. The difference in skipping probability
between the alternative and the dissimilar preview conditions was modulated by sentence frame
(b = -.46, SE = .23, t = -2.02). The separate analyses by sentence frame showed that, for highfrequency target sentence frames, there was a marginally significant difference between the
alternative and the dissimilar preview conditions, with the dissimilar preview leading to more
skips of the post-target word (b = -.33, SE = .18, z = -1.86). For low-frequency target sentence
frames, on the other hand, this effect was completely absent (z = .9)6. In summary, skipping of
the post-target word was not strongly affected by the preview manipulation.
Early processing measures: Fixation time measures
Pre-target word. There were no significant effects of preview, with the exception of a
significant interaction between the preview contrast between the identical and the alternative
preview and sentence frame in FFD (b= -14.18, SE = 5.73, t = -2.48). However, it is unclear
whether this effect can be interpreted7. There were no significant effects on other early fixation
time measures on the pre-target word (all ts< 1.96).
Target word. In the early fixation time measures we found evidence of a standard preview
benefit effect (Rayner, 1975): There was a main effect of preview in FFD and GD indicating that
fixation times on the target word were shortest in the identical condition compared to the
alternative condition (FFD: b = 15.53, SE = 4.94, t = 3.14; GD: b = 19.91, SE = 7.61, t = 2.62).
Furthermore, FFD and GD were shorter in the alternative condition than in the dissimilar
15
condition (FFD: b = 13.23, SE = 4.93, t = 2.68, GD: b = 17.34, SE = 6.86, t = 2.53), suggesting
that there was a cost of having a dissimilar, random letter preview that exceeded the cost of
having a preview that was simply a different word. On GD, this effect differed between sentence
frames (Interaction term: b = -25.47, SE = 12.59, t = -2.02). Specifically, there was a strong
difference between the alternative and the dissimilar condition in sentence frames with a highfrequency target (b = 29.12, SE = 8.43, t = 3.46) and virtually no difference in sentence frames
with a low-frequency target (b = 4.44, SE = 10.95, t = .4). Apparently, processing a lowfrequency target word was much more dependent on a correct preview than processing a highfrequency target word. The dissimilar preview proved disruptive in both cases.
Post-target word. There was a spill-over effect of the target word manipulation.
Specifically, GD on the post-target were significantly longer in the alternative preview condition
compared to the identical condition (GD: b = 22.69, SE = 7.68, t = 2.95). This effect did not
differ between sentence frames (t < 1.96).
Late processing measures
Pre-target word. There was a main effect of preview on total viewing time in those
sentence frames with a high-frequency target word, with a significant difference in TT between
the identical high-frequency target previews and the alternative low-frequency word previews (b
= 26.35, SE = 11.44, t = 2.3). There also was a significant interaction between preview
(alternative vs. identical contrast) and sentence frame (b = -35.65, SE = 16.89, t = -2.11).
Separate analyses by sentence frame (see Table 1c) showed a significant difference between the
alternative and the identical conditions in sentence frames with a high-frequency target word (b =
46.99, SE = 15.11, t = 3.11), but not in those with a low-frequency target word (b = 7.66, SE =
16.14, t = .21). Overall, preview effects for TT on the pre-target word are most likely due to
16
readers re-reading the pre-target word after having received a dissimilar preview. The fact that
such an effect only reached significance for the sentence frames with a high-frequency target
word may again indicate that processing was disrupted more when readers were expecting a
high-frequency word but receive a dissimilar preview compared to the situation in which readers
were expecting a low-frequency word and then receive a high-frequency or random letter
preview. Of course, the preceding sentence context may have influenced the effect of a dissimilar
preview in other ways as well.
Target word. The preview benefit effect we found on the earlier measures persisted in the
late measures, with longer go-past times and TT in the alternative than in the identical condition
(go-past: b = 46.29, SE = 13.2, t = 3.51; TT: b = 28.23, SE = 8.71, t = 3.24).
Post-target word. We found spillover effects on go-past time and TT (go-past: b = 64.87,
SE = 14.9, t = 4.35; TT: b = 28.67, SE = 9.81, t = 2.92), with go-past times and TTs being longer
in the alternative preview condition than in the identical condition. Again, this effect was not
modulated by sentence frame (b = -2.75, SE = 18.41, z = -.15).
Supplementary analyses
Syntactic vs. semantic preview violations. As described in the Method section, about one
third of the sentence frame/preview word combinations resulted in a semantic rather than
syntactic violation (e.g. They looked at the afternoon (sky/cut) and admired the clouds is a
semantic preview violation as opposed to the syntactically illegal preview violation in She
accidentally (cut/sky) herself while making a collage, where the first word in parentheses is the
target word and the second word in parentheses is the alternative preview). In order to determine
whether the observed effects differed by type of violation, we performed analyses with an
17
additional factor denoting whether the violation in the alternative preview condition was
syntactic (coded as 0) or semantic (coded as 1, see Appendix for details). There was no
significant interaction effect on skipping probability between violation type and preview and no
significant three-way interaction between preview, sentence frame, and violation type (all z <
1.65), suggesting that the effect we observed did not differ between violation types. There were a
small number of significant interaction effects on fixation time measures. Specifically, there was
a significant interaction between violation type and the comparison between the identical and the
alternative previews in FFD and GD on the post-target word (FFD: b= -16.67, SE = 8.24, t = 2.02; GD: b = -31.87, SE = 14.7, t = -2.17) when the alternative preview constituted a syntactic
violation, suggesting that the spillover effect from having had an alternative as opposed to an
identical preview of the target word was only present when the violation had been of a syntactic
nature. There was also a significant interaction between violation type and the comparison
between the alternative and the dissimilar previews in GD and go-past time on the target word
(GD: b = -28.42, SE = 12.97, t = -2.192; Go-past: b = -64.15, SE = 24.67, t = -2.60), indicating
that, in contrast to the main analysis, there was a difference in preview benefit between the
alternative and the dissimilar conditions, but only when the preview violation was of a syntactic
nature. However, as the present study was not designed to investigate the effect of preview
violation types and our analyses may not have enough power to detect subtle effects of violation
type, further research will be necessary in order to confirm that, for the purposes of making a
skipping decision, semantic and syntactic preview violations are equivalent (that is, that neither
type has an effect on skipping).
Content vs. function word targets and previews. Another factor that may be important in
the present study is the use of both content and function words as targets and alternative
18
previews (e.g. The old man across the street (has/hem) a very bad cold vs. My grandmother told
me she would (hem/has) that dress for me, where the first word in parentheses is the target word
and the second word in parentheses is the alternative preview). The advantage of using function
words as targets and previews is that this is likely to result in a strong syntactic violation.
However, since one of the goals of the present study was to determine whether skipping a
preview consisting of a syntactic violation occurs only for function word previews, it is
important to check whether the effects we observed were driven entirely by such target and
preview words. In order to do this, we performed analyses with an additional factor denoting
whether both target and preview words were content words (coded as 0) or whether either the
preview or target word was a function word (coded as 1, see Appendix for details); there were no
cases in which both preview and target were function words. There was no significant interaction
effect between preview and target/preview word class on any of the skipping probabilities for the
pre-target, target, and post-target words and no significant three-way interaction between
preview, sentence frame, and target/preview word class (all z < 1.96), with one exception: the
post-target word was skipped more often when the preview had been dissimilar (as opposed to
alternate) when the target word was a function word (b = .71, SE = .25, z = 2.8). Importantly, the
effect we observed on the probability of skipping the target word did not differ depending on
whether target and preview words were content or function words. There was only one
significant interaction effect on fixation time measures. Specifically, there was a significant
three-way interaction on go-past time on the target word between the identical vs. alternative
preview contrast, the sentence frame, and target word class (b = -130.11, SE = 60.08, t = -2.17),
suggesting that the preview benefit effect on go-past time on the target word was only present for
those sentence frames in which the target word was a high-frequency function word. None of
19
these interactions provide evidence that detracts from our primary claim that high frequency
parafoveal words are skipped more often than low frequency parafoveal words even if the
sentence context favors the low frequency word. Furthermore, based on these analyses, it
appears that the content or function status of a word did not modulate these effects.
Target word predictability. Finally, the target words differed in terms of their
predictability from the preceding sentence context. While the majority of the target words were
not predictable from the context, a number of target words were somewhat constrained by the
sentence context. We therefore collected target word cloze data for all sentence frames from 11
UK native speakers in order to assess predictability. Predictability was low for most, but not all
of the target words with a few exceptions (mean = .07; range = 0-.91) In order to assess whether
predictability modulated the effects we observed, we performed another set of analyses in which
we included target word predictability in the analysis. Given that there were 11 subjects, we
entered target word predictability into the model as a categorical rather than a continuous
variable (target word or target word preview were predicted by at least one norming subject,
coded as 1 –this included 29 % of items with an average predictability of .23–vs. neither target
word nor preview were predicted by any norming subjects, coded as 0—this included 71% of
items; see Appendix for details). There were no significant two- or three-way interaction effects
of target word predictability on the probability of skipping the pre-target, target, or post-target
word, suggesting that target word predictability did not affect subjects’ skipping decisions in the
present study (all z < 1.96). There were a few significant interaction effects on fixation time
measures: On the pre-target word, there were significant three-way interactions between target
word predictability, sentence frame, and each of the preview contrasts (interaction term for
identical vs. alternative: b = -34.08, SE = 13.63, t = -2.5; alternative vs. dissimilar: b = 29.94, SE
20
= 13.36, t = 2.24). This is quite interesting, as it suggests that there was a parafoveal-on-foveal
effect of preview when the target word was predictable from the sentence context, but no
parafoveal-on-foveal effect of preview when it was not predictable. Separate analyses for the
sentences with predictable target words and those with non-predictable target words produced
interactions between the preview contrasts and the sentence frame that were not significant for
those sentences in which the target word was not predictable (all ts < 1), but highly significant
for those sentences in which it was. However, this effect was only significant for the sentence
frames compatible with high-frequency target words (Interaction term for identical vs.
alternative: b = -42.02, SE = 11.43, t = -3.68; alternative vs. dissimilar: b = 30.72, SE = 11.21, t
= 2.74). Therefore, it appears that subjects may have, at least in some cases, predicted the
identity of the target word, and when the target preview was incompatible with the prediction,
they experienced some disruption. A similar effect was observed by Rayner et al. (2013). Such a
prediction mechanism could also explain the apparent parafoveal-on-foveal effects we observed
for skipping the pre-target word, although this would suggest that both prediction generation and
comparison of the prediction with the parafoveal preview can occur as much as two words ahead.
Such a suggestion might occur if the perceptual span is extended when upcoming words are
highly predictable. Finally, the interaction between the identical vs. alternative preview contrast
and word predictability on FFD on the post-target word was significant (b = -20.17, SE = 9.54, t
= 2.12), suggesting that, for predictable target words, there was no spillover effect of having had
the alternative preview—on the contrary, FFDs on the post-target word were slightly lower in the
alternative preview condition compared to the identical condition.
21
General Discussion
In the present experiment, we pitted two sources of information about the upcoming word
against each other. In the critical preview condition, the information from the available
parafoveal preview contradicted expectations about the upcoming word based on the syntactic
structure of the preceding sentence. Angele and Rayner (2012) showed that, in cases when the
preview indicates that the upcoming word is the article the, but the sentence context is
incompatible with that conclusion, the article interpretation wins: readers skip the apparent
article and incur substantial processing difficulties later on. The present experiment demonstrates
that a similar effect occurs for other three-letter words: First, random letter non-words were
skipped less often than words. Second, the frequency of the upcoming word, but not its fit with
the sentence, determined whether it was skipped. This finding is also consistent with the findings
of White (2008), who showed that high frequency words are usually skipped more often than
low-frequency words.
The present findings suggest that, at least for very short words, lexical properties of the
parafoveal preview have a substantial influence on fixation probability, while its compatibility
with the preceding sentence context matters to a far lesser degree. Additionally, properties of the
preview affect later processing, as shown by higher regression probabilities and go-past times.
Our findings also suggest that readers are able to obtain some lexical information about
parafoveal words, which fits neatly with the fact that readers seem to be able to skip even
difficult words and still comprehend a sentence without problems. On the other hand, while
expectations based on the preceding sentence context certainly play a role in shaping reading
behavior (see Bicknell & Levy, 2011), parafoveal lexical processing seems to trump the sentence
context when it comes to word skipping.
22
A remaining question is how our results relate to previous findings about the influence of
predictability on word skipping (Balota et al., 1985; Drieghe, Rayner, & Pollatsek, 2005; Ehrlich
& Rayner, 1983; Rayner & Well, 1996; for a review see Brysbaert, Drieghe, & Vitu, 2005),
which suggest that readers are more likely to skip highly predictable words. Given this, why did
we not see more pronounced effects of predictability in the present study? As noted above, this
likely occurred because our target words were, in general, not strongly predictable (see Appendix
for details). It also remains to be determined whether our results generalize to longer words (for a
recent study on the skipping of longer words, see Choi & Gordon, 2013).
In summary, the present study demonstrates that skipping of short words is strongly
influenced by the frequency of its parafoveal preview. This suggests that the the-skipping effects
observed by O’Regan, (1979), Gautier et al. (2000), and Angele and Rayner (2012) are not
necessarily specific to the definite article the, but rather, they occur as a function of its word
frequency. On a more general level, our results underline the importance of parafoveal
orthographic and lexical processing in comparison with higher-level processing such as syntactic
or semantic processing. While syntactic integration and semantic processing occur during or after
a word has been fixated, the information available about a word that is still in the parafovea is
mainly of either an orthographic or lexical nature. Importantly, this does not mean that
parafoveal information about a word is irrelevant once the eyes have moved—on the contrary,
we found clear effects of the preview of a word having been dissimilar once that word is finally
fixated and even on the subsequent fixations. Clearly, however, there is no evidence with this
experimental paradigm that syntactic and semantic information about the upcoming word and its
fit with the sentence affect the timing of the skipping decision.
Implications for computational models
23
Our findings show that readers routinely skip parafoveal high frequency words without
taking the syntactic or semantic sentence context into account. This demonstrates that, at least
with regards to skipping short words, the current implementations of both the E-Z Reader and
SWIFT models are adequate: In both models, the time needed to parafoveally process a word
(and its fixation probability) is strongly dependent on its frequency. Both models also predict that
the processing time of a word should be influenced by its predictability, however, this is where
our results do not correspond to the model predictions, as we saw no effect of fit with the
sentence context. It is important to note, however, that both E-Z Reader 9 (Pollatsek, Reichle, &
Rayner, 2006), and SWIFT focus on word identification rather than semantic integration. As
such, they both allow parafoveal identification of an upcoming word prior to skipping it. Version
10 of E-Z Reader (Reichle et al., 2009) goes further than this by including a processing stage
representing the syntactic and semantic integration of a word into the sentence context. If one
assumes that the integration stage in E-Z Reader 10 would necessarily fail to integrate the
infelicitous the preview into the sentence, E-Z Reader 10 might be able to explain why there was
disruption only subsequent to skipping the infelicitous the preview. Keeping in mind that there
are previous results supporting an effect of predictability on word skipping and that our
experiment really only measures the difference in skipping between an unpredictable and
grammatically illegal (or, in the case of the semantic violations, at least implausible, if not
anomalous) parafoveal word, some changes in how the models treat the predictability of
parafoveal words might be necessary. However, such changes would be rather small
modifications of existing mechanisms in both models and would not require that new
mechanisms be proposed. Furthermore, these effects can be accounted for by both parallel and
serial accounts of word identification.
24
The unexpected parafoveal-on-foveal effect of the preview condition on skipping
probability and fixation time on the pre-target word might have stronger implications for the
models. If there is indeed an early parafoveal plausibility check, such a mechanism is not
present in either model and would have to be added. One might argue that processing far ahead
of the fixated word may be more in the “spirit” of parallel processing, however, it would not be
impossible to conceive of a serial model that involves a very early parallel visual stage in which
all letters in the perceptual span are superficially processed at the same time (in fact, the “V”
stage in E-Z Reader is described as exactly such a process, though this is not implemented in
current versions of the model). This superficial processing could involve checking whether
parafoveal word information is consistent with those words that are pre-activated from
processing the context. It is important to note at this point that, in order to warrant model
modifications, the parafoveal-on-foveal effects described above will have to be replicated in a
dedicated experiment. They do suggest some interesting directions for future research, however.
In summary, our finding that high frequency parafoveal words are often skipped without
taking the sentence context into account is, in principle, compatible with both E-Z Reader and
SWIFT (pending minor modifications). Explaining our findings on the pre-target word might
require more fundamental model changes, but further research is required to replicate those
effects in a controlled experiment. On a more general level, our results indicate that there is at
least one important process during reading which does not seem to be affected by word
predictability from the context, but just by the properties of the parafoveal word. This may pose a
problem to models, such as that of Bicknell and Levy (2010), that propose that readers constantly
evaluate the available evidence about previously fixated, currently fixated, and upcoming words
since such an evaluation should alert readers to the mismatch between prediction and actual
25
parafoveal information. While we did find some evidence for such a parafoveal plausibility
check, the effects we observed were not strong enough to warrant the assumption that such a
check is carried out during every fixation. Rather, it seems that, during normal reading,
parafoveal plausibility is only checked once in a while if at all.
26
References
Angele, B., & Rayner, K. (2013). Processing the in the parafovea: Are articles
skipped automatically? Journal of Experimental Psychology: Learning, Memory, and
Cognition, 39(2), 649-662. doi:10.1037/a0029294
Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-effects modeling
with crossed random effects for subjects and items. Journal of Memory and Language,
59, 390–412. doi:10.1016/j.jml.2007.12.005
Balota, D. A., Pollatsek, A., & Rayner, K. (1985). The interaction of contextual
constraints and parafoveal visual information in reading. Cognitive Psychology, 17, 364–
390.
Bates, D., Maechler, M., Bolker, B. & Walker S. (2013). lme4: Linear mixedeffects models using Eigen and S4. R package version 1.0-5.
http://CRAN.R-project.org/package=lme4
Bicknell, K., & Levy, R. (2010). A rational model of eye movement control in
reading. In Proceedings of the 48th Annual Meeting of the Association for Computational
Linguistics (pp. 1168–1178).
Bicknell, K., & Levy, R. (2011). Why readers regress to previous words: A
statistical analysis. In L. Carlson, C. Hölscher, & T. Shipley (Eds.), Proceedings of the
33rd Annual Conference of the Cognitive Science Society. (pp.
931–936). Austin, TX: Cognitive Science Society.
27
Brysbaert, M., Drieghe, D., & Vitu, F. (2005). Word skipping: Implications for
theories of eye movement control in reading. In G. Underwood (Ed.), Cognitive processes
in eye guidance (pp. 53-77). Oxford: Oxford University Press
Choi, W., & Gordon, P. C. (2013). Coordination of word recognition and
oculomotor control during reading: the role of implicit lexical decisions. Journal of
Experimental Psychology: Human Perception and Performance, 39(4), 1032–1046.
doi:10.1037/a0030432
Drieghe, D., Rayner, K., & Pollatsek, A. (2005). Eye movements and word
skipping during reading revisited. Journal of Experimental Psychology: Human
Perception and Performance, 31, 954-969.
Davis, C. J. (2005). N-watch: A program for deriving neighborhood size and other
psycholinguistic statistics. Behavior Research Methods, 37(1), 65-70.
Ehrlich, S. F., & Rayner, K. (1981). Contextual effects on word perception and
eye movements during reading. Journal of Verbal Learning and Verbal Behavior, 20,
641–655. doi:10.1016/S0022-5371(81)90220-6
Engbert, R., Nuthmann, A., Richter, E. M., & Kliegl, R. (2005). SWIFT: A
dynamical model of saccade generation during reading. Psychological Review, 112, 777–
813. doi:10.1037/0033-295X.112.4.777
Gautier, V., O’Regan, J. K., & Le Gargasson, J. F. (2000). “The-skipping”
revisited in French: programming saccades to skip the article “les”. Vision Research, 40,
2517–2531.
28
Hautala, J., Hyönä, J., & Aro, M. (2011). Dissociating spatial and letter-based
word length effects observed in readers’ eye movement patterns. Vision Research, 51,
1719–1727. doi:10.1016/j.visres.2011.05.015
Kennedy, A. (1998). The influence of parafoveal words on foveal inspection time:
Evidence for a processing trade-off. In G. Underwood (Ed.), Eye Guidance in Reading
and Scene Perception (pp. 149–179). Oxford, UK: Elsevier.
Kennedy, A., Murray, W., & Boissiere, C. (2004). Parafoveal pragmatics revisited.
European Journal of Cognitive Psychology, 16, 128–153.
Kennedy, A., & Pynte, J. (2005). Parafoveal-on-foveal effects in normal reading.
Vision Research, 45, 153–168.
Kennedy, A., Pynte, J., & Ducrot, S. (2002). Parafoveal-on-foveal interactions in
word recognition. The Quarterly Journal of Experimental Psychology Section A, 55,
1307–1337.
Kliegl, R., Nuthmann, A., & Engbert, R. (2006). Tracking the mind during
reading: The influence of past, present, and future words on fixation durations. Journal of
Experimental Psychology-General, 135, 12–35.
Levy, R., Bicknell, K., Slattery, T., & Rayner, K. (2009). Eye movement evidence
that readers maintain and act on uncertainty about past linguistic input. Proceedings of
the National Academy of Sciences, 106, 21086–21090. doi:10.1073/pnas.0907664106
Liversedge, S. P., & Findlay, J. M. (2000). Saccadic eye movements and
cognition. Trends in Cognitive Sciences, 4, 6–14. doi:10.1016/S1364-6613(99)01418-7
McConkie, G. W., & Rayner, K. (1975). The span of the effective stimulus during
a fixation in reading. Perception & Psychophysics, 17, 578–587.
29
O’Regan, K. (1979). Saccade size control in reading: Evidence for the linguistic
control hypothesis. Perception & Psychophysics, 25, 501–509. doi:10.3758/BF03213829
Pollatsek, A., Reichle, E.D., & Rayner, K. (2006). Tests of the E-Z Reader model:
Exploring the interface between cognition and eye-movement control. Cognitive
Psychology, 52, 1-56.
R Development Core Team, R. (2013). R: a language and environment for
statistical computing. R Foundation for Statistical Computing.
Rayner, K. (1975). The perceptual span and peripheral cues in reading. Cognitive
Psychology, 7, 65–81. doi:10.1016/0010-0285(75)90005-5
Rayner, K. (1998). Eye movements in reading and information processing: 20
years of research. Psychological Bulletin, 124, 372–422. doi:10.1037/00332909.124.3.372
Rayner, K. (2009). The Thirty-Fifth Sir Frederick Bartlett Lecture: Eye
movements and attention in reading, scene perception, and visual search. The Quarterly
Journal of Experimental Psychology, 62, 1457–1506. doi:10.1080/17470210902816461
Rayner, K., Angele, B., Schotter, E. R. and Bicknell, K. (2013). On the processing
of canonical word order during eye fixations in reading: Do readers process transposed
word previews? Visual Cognition, 21 (3), 353-381.
Rayner, K., & Liversedge, S. (2011). Linguistic and cognitive influences on eye
movements during reading. In S. Liversedge, I. Gilchrist, & S. Everling (Eds.), The
Oxford Handbook of Eye Movements. Oxford, UK: Oxford University Press.
Rayner, K., & McConkie, G. W. (1976). What guides a reader’s eye movements?
Vision Research, 16, 829–837. doi:10.1016/0042-6989(76)90143-7
30
Rayner, K., & Well, A. D. (1996). Effects of contextual constraint on eye
movements in reading: A further examination. Psychonomic Bulletin & Review, 3, 504–
509.
Reichle, E. D., Pollatsek, A., Fisher, D. L., & Rayner, K. (1998). Toward a model
of eye movement control in reading. Psychological Review, 105, 125–157.
doi:10.1037/0033-295X.105.1.125
Reichle, E. D., Warren, T., & McConnell, K. (2009). Using E-Z Reader to model
the effects of higher level language processing on eye movements during reading.
Psychonomic Bulletin & Review, 16, 1–21. doi:10.3758/PBR.16.1.1
Schotter, E.R., Angele, B., & Rayner, K. (2012). Parafoveal processing in reading.
Attention, Perception, & Psychophysics, 74, 5–35. doi:10.3758/s13414-011-0219-2
Slattery, T. J., Angele, B., & Rayner, K. (2011). Eye movements and display
change detection during reading. Journal of Experimental Psychology. Human Perception
and Performance, 37, 1924–1938. doi:10.1037/a0024322
Vitu, F., O’Regan, J. K., Inhoff, A. W., & Topolski, R. (1995). Mindless reading:
Eye-movement characteristics are similar in scanning letter strings and reading texts.
Perception & Psychophysics, 57(3), 352–364. doi:10.3758/BF03213060
White, S. J. (2008). Eye movement control during reading: Effects of word
frequency and orthographic familiarity. Journal of Experimental Psychology: Human
Perception and Performance, 34, 205–223. doi:10.1037/0096-1523.34.1.205
White, S., Rayner, K., & Liversedge, S. (2005). Eye movements and the
modulation of parafoveal processing by foveal processing difficulty: A reexamination.
Psychonomic Bulletin & Review, 12, 891–896. doi:10.3758/BF03196782
31
Acknowledgments
This research was supported by Leverhulme Trust Grant F/00 180/AN, Economic and
Social Research Council Grant ES/I032398/1 and by National Institutes of Health Grant
HD26765. We thank Emily Morgan and Roger Levy for their help with collecting the cloze
norming data, and Reinhold Kliegl for comments on a prior draft.
32
Footnotes
1
An extreme position on this question would hold that skipping is almost entirely
explainable by oculomotor factors, including word length (for an example, see Vitu, O’Reagan,
Inhoff, & Topolski, 1995). However, this fails to explain how word frequency and predictability
can have an effect on word skipping as detailed below.
2
In rare cases, the oculomotor error built into E-Z Reader can lead to accidental skipping
of a word. However, in this case, skipping the word would not be considered intentional.
3
At a refresh rate of 150 Hz, the display changes took an average of 3 ms and a
maximum of 6 ms to be completed after they were initiated. As a consequence, after removing
those trials with very late display changes (see Results and Discussion section), the average
display change finished less than 2ms after fixation onset.
4
There is evidence that awareness of display changes can influence some eye-movement
measures (White, Rayner, & Liversedge, 2005). Out of 53 subjects that were originally tested,
the data of 18 subjects who noted display changes on their own without prompting or noticed
more than 5 changes were excluded from the analysis because of this (leaving a total of 35
subjects included in the analysis). An analysis of the data from the 18 discarded subjects showed
few differences in terms of the effects pattern compared to the subjects in the main analysis;
however, a number of effects failed to reach significance, likely due to power limitations. The
skipping probability effects that did not reach significance for the discarded subjects were the
33
effect of preview (both contrasts) on the probability of fixating the pre-target word, the
difference between the dissimilar and the alternative conditions in terms of the probability of
skipping the target word as well as its interaction with sentence frame, and the effect of the
interaction between the difference between the dissimilar and the alternative preview and the
sentence frame on the probability of skipping the post-target word. The fixation time effects that
did not reach significance for the discarded subjects were the interaction between preview and
sentence frame on FFD on the pre-target word, the difference between the dissimilar and the
alternative preview on FFD and GD on the target word, the interaction between preview and
sentence frame on GD on the target word, the effect of sentence frame on FFD on the post-target
word, on GD on the target word, and on go-past time on the post-target word, and, finally, the
difference between the identical and alternative preview conditions on GD and go-past time on
the post-target word. There was only one effect that was significant for the discarded subjects but
not for the included subjects, namely an interaction between preview and sentence frame on GD
on the post-target word (b = -76.82, SE = 26.19, t = -2.93).
5
Random slopes for the interaction term could not be included, as the majority of the
models no longer converged in this case. A small minority of models did not converge even with
the restricted random effects specification described above. In this case, we report a more
restricted model including random slopes only for the preview effect.
6
There was an overall effect of preview on the probability of making a regression out of
the post-target word, with more regressions in the alternative preview condition than in the
34
identical preview condition (b = .58, SE = .19, z = 3.11). This effect was not modulated by
sentence frame and suggests that having had a parafoveal preview of a target word (or a
dissimilar, random letter nonword) that does not fit in the later sentence context does disrupt
further processing to some degree. Even though readers have seen the correct target word at this
point, the preview information still seems to have some effect on their processing, possibly at the
integration level.
7
We are very cautious in interpreting this interaction effect, which is limited to FFD, for
three reasons: (1) in order to detect the syntactic violation caused by the mismatch of sentence
frame and preview, readers would have to have fully identified the preview and integrated it into
the sentence frame before actually fixating the target word and (2) even if they did this, it is not
clear why fixations would then be shorter in the syntactic violation condition rather than longer.
Furthermore, (3) it is unclear why the syntactic violation would not also be detected in the highfrequency sentence frames as well.
35
Table 1a: Condition means on the pre-target word.
Fixation time measures
Preview
Identical
Probabilities
First fixation duration Gaze duration Go-past time Total viewing time Skipping probability Probability of regressions out
221 (57)
248 (94.7)
272 (133)
292 (153)
0.17 (0.38)
0.05 (0.22)
high Alternative
223 (57)
254 (104)
286 (173)
334 (193)
0.15(0.36)
0.06(0.24)
Dissimilar
224 (58)
259 (112)
297 (192)
337 (207)
0.11 (0.31)
0.09(0.28)
Identical
227 (65)
261 (106)
307 (191)
319 (176)
0.12(0.32)
0.09 (0.2.9)
low Alternative
218 (56)
253 (114)
319 (254)
334 (217)
0.14(0.34)
0.1(0.3)
Dissimilar
224 (59)
259 (125)
305 (218)
329 (183)
0.14(0.35)
0.08 (0.27)
Standard deviations are in parentheses.
Table 1b: LMM analyses on the pre-target word. Each column represents a model.
First fixation
duration
Preview
Sentence frame
Interactions
B
SE
Gaze duration
t
b
SE
Go-past time
t
b
291.49
SE
Total viewing time
t
b
SE
t
Skipping
probability
b
SE
z
(Intercept)
220.33
4.02
54.74 250.40
7.40 33.83
12.96 22.50
318.70
14.79 21.55 -2.57
0.22 -11.84
Alternative vs. Identical
-2.85
3.40
-0.84
-1.05
5.30
-0.20
9.50
10.75
0.88
26.35
11.44
2.30
0.45
0.18
2.49
Dissimilar vs. Alternative
2.70
3.07
0.88
4.89
5.52
0.89
0.82
12.09
0.07
2.91
11.46
0.25
-0.44
0.19
-2.30
6.02
High vs. Low
1.91
3.21
0.60
6.27
1.04
23.98
10.70
2.24
9.29
9.52
0.98
-0.30
0.17
-1.72
Alternative vs. Identical ×
Sentence frame
-14.18
5.73
-2.48
-17.21
10.46 -1.65
-10.10
18.81
-0.54
-35.65
16.89
-2.11
0.58
0.32
1.81
Dissimilar vs. Alternative ×
Sentence frame
5.72
5.64
1.01
1.01
10.36 0.10
-23.88
18.67
-1.28
-0.75
16.76
-0.04
0.47
0.31
1.49
b: Regression coefficient, SE: standard error, t or z: test statistic (b/SE). Cells with |t| or |z| >= 1.96 are marked in bold.
36
Table 1c: LMM analyses on the pre-target word by sentence frame. Each column represents a model.
Sentence frames
First fixation duration
with high-frequency target words only
Preview
B
SE
t
Sentence frames
First fixation duration
with low-frequency target words only
Total viewing time
b
SE
t
(Intercept)
220.30
4.25
51.84 316.33 15.16 20.87
Alternative vs. Identical
3.48
4.84
0.72
46.99 15.11 3.11
Dissimilar vs. Alternative
-0.28
4.08
-0.07
3.27
16.83 0.19
Preview
b
SE
t
Total viewing time
b
SE
t
(Intercept)
221.12
4.30
51.39 322.38 15.95 20.22
Alternative vs. Identical
-9.89
4.27
-2.32
7.66
16.14 0.47
Dissimilar vs. Alternative
5.33
4.17
1.28
3.12
14.57 0.21
b: Regression coefficient, SE: standard error, t or z: test statistic (b/SE). Cells with |t| or |z| >= 1.96 are marked in bold.
37
Table 2a: Condition means on the target word.
Fixation time measures
Preview
Identical
Probabilities
First fixation duration Gaze duration Go-past time Total viewing time Skipping probability Probability of regressions out
235 (73)
250 (95)
283 (162)
275 (143)
0.52 (0.5)
0.05 (0.22)
high Alternative
244 (67)
261 (95)
327 (192)
303 (155)
0.47 (0.5)
0.08 (0.28)
Dissimilar
272 (80)
294 (99)
354 (199)
321 (141)
0.46 (0.5)
0.09(0.28)
241 (69)
270 (107)
306 (165)
305 (156)
0.49(0.5)
0.048 (0.21)
low Alternative
Identical
259 (80)
296 (112)
355 (211)
334 (166)
0.54(0.5)
0.07 (0.26)
Dissimilar
266 (77)
300 (115)
354 (194)
332 (161)
0.41(0.49)
0.08 (0.28)
Standard deviations are in parentheses.
Table 2b: LMM analyses on the target word. Each column represents a model.
First fixation duration
Preview
Sentence frame
Interactions
B
SE
(Intercept)
251.49
6.61
t
Gaze duration
b
SE
t
Go-past time
b
SE
Total viewing time
t
B
SE
t
Skipping probability
b
38.03 277.25 8.42 32.91 328.40 10.48 31.33 305.87 11.27 27.13 -0.03
SE
z
0.15
-0.19
Alternative vs. Identical
15.53
4.94
3.14
19.91
7.61
2.62
46.29 13.20 3.51
28.23
8.71
3.24
-0.01
0.10
-0.08
Dissimilar vs. Alternative
13.23
4.93
2.68
17.34
6.86
2.53
12.51 14.71 0.85
7.77
8.43
0.92
-0.31
0.11
-2.72
High vs. Low
8.15
4.44
1.84
24.29
6.44
3.77
16.54 11.28 1.47
28.96
6.97
4.16
0.00
0.10
0.04
-2.36
-5.52
Alternative vs. Identical × Sentence frame
6.35
9.25
0.69
11.79 13.00 0.91
17.38 -0.32
0.52
0.21
2.50
Dissimilar vs. Alternative × Sentence frame
-17.20
9.01
-1.91
-25.47 12.59 -2.02 -21.08 24.19 -0.87 -13.89 16.53 -0.84
24.90 -0.09
-0.61
0.21
-2.91
b: Regression coefficient, SE: standard error, t or z: test statistic (b/SE). Cells with |t| or |z| >= 1.96 are marked in bold.
38
Table 2c: LMM analyses on the target word by sentence frame. Each column represents a model.
Sentence frames with
high-frequency target words only
Gaze duration
SE
z
266.51 8.19 32.56 -0.03
0.17
-0.18
Alternative vs. Identical
12.77 8.60 1.49
-0.29
0.17
-1.69
Dissimilar vs. Alternative
29.12 8.43 3.46
0.00
0.18
-0.02
b
(Intercept)
Preview
SE
Skipping probability
t
b
Sentence frames with
low-frequency target words only
Gaze duration
SE
z
289.40 9.96 29.06 -0.07
0.14
-0.52
25.52 11.73 2.18
0.25
0.15
1.64
4.44
-0.62
0.15
-4.17
b
(Intercept)
Preview
Alternative vs. Identical
Dissimilar vs. Alternative
SE
Skipping probability
t
10.95 0.41
b: Regression coefficient, SE: standard error, t or z: test statistic (b/SE). Cells with |t| or |z| >= 1.96 are marked in bold.
b
39
Table 3a: Condition means on the post-target word.
Fixation time measures
Preview
Identical
Probabilities
First fixation duration Gaze duration Go-past time Total viewing time Skipping probability Probability of regressions out
228 (69)
255 (106)
314 (199)
297 (154)
0.351 (0.48)
0.09(0.28)
high Alternative
235 (71)
275 (130)
381 (268)
326 (187)
0.324 (0.47)
0.15 (0.36)
Dissimilar
226 (70)
271 (128)
386 (264)
335 (179)
0.279 (0.45)
0.17(0.38)
242 (76)
283 (139)
357 (217)
331 (180)
0.336 (0.47)
0.11 (0.31)
low Alternative
Identical
240 (76)
305 (140)
424 (259)
355 (180)
0.306 (0.46)
0.163 (0.37)
Dissimilar
238 (75)
296 (137)
405 (254)
360 (187)
0.34 (0.47)
0.156 (0.364)
Standard deviations are in parentheses.
Table 3b: LMM analyses on the post-target word. Each column represents a model.
First fixation duration*
(Intercept)
Preview
Sentence frame
Interactions
Gaze duration
233.84
5.93
39.46
276.49 9.39 29.45 374.59 14.06 26.64 328.42 11.95 27.49 -0.92
b
t
B
SE
Skipping probability Probability of regressions out*
t
t
SE
Total viewing time
SE
b
SE
Go-past time
B
t
b
SE
z
b
SE
z
0.11
-8.78
-2.09
0.14
-14.55
Alternative vs. Identical
4.26
4.28
1.00
22.69
7.68
2.95
64.87 14.90 4.35
28.67
9.81
2.92
-0.14
0.11
-1.22
0.58
0.19
3.11
Dissimilar vs. Alternative
-4.54
4.16
-1.09
-5.09
7.78
-0.65
-5.76
7.04
9.70
0.73
-0.04
0.13
-0.33
0.15
0.14
1.07
High vs. Low
11.20
3.27
3.43
26.50
8.16
3.25
33.63 16.05 2.10
28.32 10.45 2.71
-0.05
0.14
-0.37
0.12
0.12
0.98
Alternative vs. Identical × Sentence frame
-6.91
8.06
-0.86
6.52
14.05 0.46
-5.71
27.12 -0.21
-2.75
18.41 -0.15
0.01
0.22
0.05
-0.23
0.32
-0.72
Dissimilar vs. Alternative × Sentence frame
6.23
7.92
0.79
-2.17
13.80 -0.16 -20.04 26.89 -0.75
-0.11
18.05 -0.01
0.46
0.23
2.02
-0.16
0.28
-0.57
16.57 -0.35
b: Regression coefficient, SE: standard error, t or z: test statistic (b/SE). Cells with |t| or |z| >= 1.96 are marked in bold.
40
Table 3c: LMM analyses on the post-target word by sentence frame. Each column represents a model.
Sentence frames with
Skipping probability
high-frequency target words only
Preview
b
SE
z
(Intercept)
-0.91
0.11
-8.11
Alternative vs. Identical
-0.11
0.16
-0.72
Dissimilar vs. Alternative
-0.33
0.18
-1.86
Sentence frames with
Skipping probability
low-frequency target words only
Preview
b
SE
z
(Intercept)
-0.94
0.14
-6.94
Alternative vs. Identical
-0.13
0.17
-0.77
Dissimilar vs. Alternative
0.15
0.17
0.90
b: Regression coefficient, SE: standard error, t or z: test statistic (b/SE). Cells with |t| or |z| >= 1.96 are marked in bold.
41
Figure captions
1.
Example stimuli from the present study. While readers were fixating to the left of the invisible boundary (dashed lined), the target
word in each of the sentence frames (dog/dim) was replaced with a preview as shown in the left column. After a reader crossed the boundary,
the preview was replaced with the actual target word (right column).
42
Figure 1.
Sentence frame
High-frequency
target
Low-frequency
target
Preview
High-frequency
Low-frequency
Random letters
High-frequency
Low-frequency
Random letters
Example sentence before display change
Example sentence after display change
The excitable dog was eager to go for his walk.
The excitable dim was eager to go for his walk.
The excitable dog was eager to go for his walk. The excitable hev was eager to go for his walk.
The increasingly dog light made it hard to see.
The increasingly dim light made it hard to see.
The increasingly bup light made it hard to see.
The increasingly dim light made it hard to see.
43
Appendix
Table A1: List of sentences used in the experiment along with predictability, violation, and target and preview word class
criteria that were used for the supplementary analyses. Parentheses mark the target word and the alternative preview word.
Note that each subject only saw one of the two sentences in a pair.
Pair
1
1
2
2
3
3
4
4
5
5
6
6
7
7
8
Sentence
Number
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Sentence
Last year, he lost (all/ant) his money on the stock market.
There was a massive (ant/all) infestation in the old house.
Today you must (buy/hag) the cake for the birthday party.
In his dream, he saw an ancient (hag/buy) who scared him.
My brother has terrible (aim/ape) and he will never play baseball.
This weekend, I saw a large (ape/aim) and her new-born baby at the zoo.
Every Sunday, the little (boy/beg) would go to church with his family.
The children would always (beg/boy) their mother for ice cream.
She tried with all her might (but/mat) the heavy bookshelf would not budge.
The karate teacher threw his students down on the large (mat/but) multiple times.
Using the new system, the students (can/cow) access their grades online.
Early in the morning, he walked his only (cow/can) out to pasture.
It was an emotional (day/dew) for all of the family members.
They noticed the moist (dew/day) on the rose's petals.
They attempted to keep their feet (dry/duo) while crossing the small stream.
Cloze
Predictability
0.13
0.00
0.00
0.00
0.00
0.00
0.36
0.00
0.27
0.91
0.09
0.00
0.27
0.09
0.27
Violation
syntactic
syntactic
syntactic
syntactic
syntactic
semantic
syntactic
syntactic
syntactic
syntactic
syntactic
semantic
semantic
semantic
syntactic
Target
word
class
closed
open
open
open
open
open
open
open
closed
open
closed
open
open
open
open
Preview
word
class
open
closed
open
open
open
open
open
open
open
closed
open
open
open
open
open
44
Pair
8
9
9
10
10
11
11
12
12
13
13
14
14
15
15
16
16
17
17
18
18
19
19
20
20
21
21
22
22
Sentence
Number
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
Sentence
The famous singing (duo/dry) performed for the entranced audience.
The manager thought it would be great if everyone (did/don) more work in the office.
He would never (don/did) the red hat for the formal ceremony.
The excitable (dog/dim) was anxious to go for his walk.
The increasingly (dim/dog) light made it hard to see.
The liked the look of the large (cup/nip) that was for sale in the store.
My cat will sometimes (nip/cup) my hand when it is playful.
Jane rinsed her left (eye/emu) because something was in it.
With ruffled feathers, the angry (emu/eye) raced up and down his paddock.
The house was very (far/cap) from his school.
His dirty (cap/far) did not make John look any better.
After the long drive, we had to stop (for/rub) gas and supplies.
Her friend gave Maria a back (rub/for) and she felt much better.
The ranger reported seeing (few/fin) birds this year.
The fish was left with only a single (fin/few) after the attack.
We filled our tank with (gas/shy) and then drove off into the night.
The boy was very (shy/gas) when he performed on stage.
She requested that I immediately (get/gym) her a chocolate bar from the store.
I promised myself that I would go to the local (gym/get) every day this week.
The man asked the tribal (god/gag) for relief from the persistent rain.
The burglar will (gag/god) the frightened old lady.
She told her husband that they (had/oar) to pay the bills by the end of the month.
The man picked up the extra (oar/had) and prepared to row across the lake.
The old man across the street (has/hem) a very bad cold.
My grandmother told me she would (hem/has) that dress for me.
That night, she asked (her/hit) father if she could go to the party.
The young child (hit/her) the ball over the fence.
The woman asked (him/bug) if he knew where the High Street was.
During our road trip a huge (bug/him) splattered onto our window.
Cloze
Predictability
0.00
0.00
0.00
0.18
0.00
0.00
0.00
0.18
0.00
0.00
0.00
0.36
0.27
0.00
0.09
0.18
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.09
0.00
0.09
0.00
Violation
syntactic
syntactic
syntactic
syntactic
syntactic
semantic
semantic
semantic
semantic
syntactic
syntactic
syntactic
syntactic
syntactic
semantic
syntactic
syntactic
syntactic
syntactic
semantic
syntactic
syntactic
syntactic
syntactic
syntactic
syntactic
syntactic
syntactic
syntactic
Target
word
class
open
open
open
open
open
open
open
open
open
open
open
closed
open
closed
open
open
open
open
open
open
open
closed
open
closed
open
open
open
closed
open
Preview
word
class
open
open
open
open
open
open
open
open
open
open
open
open
closed
open
closed
open
open
open
open
open
open
open
closed
open
closed
open
open
open
closed
45
Pair
23
23
24
24
25
25
26
26
27
27
28
28
29
29
30
30
31
31
32
32
33
33
34
34
35
35
36
36
37
Sentence
Number
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
Sentence
The room was incredibly (hot/hen) and the men decided to leave.
The mother (hen/hot) watched over her chicks in the coop.
The girl knew (how/hid) the magician's trick worked.
The little girl (hid/how) where no one could find her.
The puppy devoured (its/lip) dinner in a few short minutes.
Mary's expensive (lip/its) treatment turned out to have no effect at all.
The office workers regularly (lay/rum) down the papers in just the right order.
Jacob enjoyed having (rum/lay) with his coffee.
I have never (met/pad) a celebrity before.
She liked standing on the soft (pad/met) in the bedroom.
Every year they travel somewhere (new/nap) like Greece or Brazil.
My dog loves taking a long (nap/new) when its hot outside.
My baby cousin will not eat spinach (nor/net) broccoli but enjoys eating pizza.
The old man pulled in his heavy (net/nor) and discovered that he had caught a swordfish.
They want to leave (now/hog) rather than later in the evening.
The husband would (hog/now) all of the blankets at night.
The volunteers wiped the toxic (oil/ode) off of the animals as best they could.
The children wrote a beautiful (ode/oil) to their teachers.
The manuscript was very (old/owl) and the team had to handle it carefully.
I saw a large (owl/old) flying over the barn.
There was only (one/rid) doll to play with.
He could finally (rid/one) himself of all the old paperwork.
We walked out onto (our/orb) front porch with a cup of coffee.
A bright shimmering (orb/our) appeared in the night sky.
She could probably (pay/pea) for the couch with her credit card.
She would not eat even a single (pea/pay) off her plate.
They had to pay fifty pounds (per/nod) person for tickets to the game.
The nice man would kindly (nod/per) every time someone entered the store.
Each student must (put/pew) their chairs on their desk at the end of the day.
Cloze
Predictability
0.09
0.09
0.00
0.00
0.00
0.00
0.00
0.00
0.09
0.00
0.45
0.18
0.00
0.09
0.09
0.00
0.00
0.00
0.27
0.00
0.55
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
Violation
syntactic
syntactic
syntactic
syntactic
semantic
syntactic
semantic
semantic
syntactic
syntactic
syntactic
semantic
syntactic
syntactic
syntactic
semantic
semantic
semantic
syntactic
semantic
syntactic
syntactic
syntactic
syntactic
syntactic
semantic
syntactic
syntactic
syntactic
Target
word
class
open
open
closed
open
closed
open
open
open
open
open
open
open
closed
open
open
open
open
open
open
open
open
open
closed
open
open
open
closed
open
open
Preview
word
class
open
open
open
closed
open
closed
open
open
open
open
open
open
open
closed
open
open
open
open
open
open
open
open
open
closed
open
open
open
closed
open
46
Pair
37
38
38
39
39
40
40
41
41
42
42
43
43
44
44
45
45
46
46
47
47
48
48
49
49
50
50
51
51
Sentence
Number
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
Sentence
The caretaker bought an expensive (pew/put) for the church.
The couple (own/sap) a large house in the country.
The sweet (sap/own) of some trees can be used for cooking.
There were only (six/rot) men in the hunting party.
The vegetables began to immediately (rot/six) after she had bought them.
They started the long and bloody (war/spy) over a silly disagreement.
The man would constantly (spy/war) on his neighbors.
Maybe today (you/rug) should clean the living room.
I spilled juice on the precious (rug/you) when she bumped my arm.
The best team will (win/soy) the championship.
She drinks (soy/win) milk only because she is lactose intolerant.
Jonathan walked (two/tug) miles yesterday.
She had to forcefully (tug/two) the shirt to get it out from under the dresser.
The little girl asked (why/wig) the sky is blue.
I wore my purple (wig/why) the other day for the fancy dress party.
Jane stated (who/wad) had robbed her.
She had a giant (wad/who) of notes in her wallet.
Despite the bad weather, John (was/tee) still planning to go to the concert.
She placed the ball on the small (tee/was) and hit it into the distance.
I asked her to send me the package (via/tub) air mail as soon as possible.
I like to relax in my large (tub/via) at the end of a long day.
To her dismay, Susan still cannot (use/web) the heavy weights at the gym.
After her intricate (web/use) was destroyed, the spider was very confused.
I like to make (tea/den) on hot summer days.
I saw the little foxes near the hidden (den/tea) when I was walking in the forest.
During dinner my father (sat/tip) at the head of the table.
I left a large (tip/sat) for the helpful waiter.
I was annoyed by the hefty (tax/urn) I had to pay.
They noticed the small (urn/tax) in the funeral director's office.
Cloze
Predictability
0.00
0.00
0.00
0.00
0.18
0.18
0.00
0.09
0.18
0.27
0.00
0.00
0.00
0.09
0.00
0.00
0.09
0.18
0.00
0.00
0.00
0.00
0.00
0.00
0.36
0.00
0.09
0.00
0.00
Violation
syntactic
semantic
semantic
syntactic
syntactic
semantic
semantic
syntactic
semantic
syntactic
semantic
syntactic
syntactic
syntactic
syntactic
syntactic
syntactic
syntactic
syntactic
syntactic
syntactic
syntactic
syntactic
semantic
semantic
syntactic
syntactic
semantic
semantic
Target
word
class
open
open
open
closed
open
open
open
closed
open
open
open
closed
open
closed
open
closed
open
closed
open
closed
open
open
open
open
open
open
open
open
open
Preview
word
class
open
open
open
open
closed
open
open
open
closed
open
open
open
closed
open
closed
open
closed
open
closed
open
closed
open
open
open
open
open
open
open
open
47
Pair
52
52
53
53
54
54
55
55
56
56
57
57
58
58
59
59
60
60
61
61
62
62
63
63
64
64
65
65
66
Sentence
Number
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
Sentence
She asked me if there (are/pup) any doughnuts left over.
The campers stayed away from the small (pup/are) because they knew its mother was near.
Air traffic controllers must sometimes (act/lab) quickly in order to avoid a tragedy.
The student worked in the secret (lab/act) every weekend.
The man provided (aid/awe) to the woman before it was too late.
The critics expressed (awe/aid) at the orchestra's inspiring rendition of Beethoven's fifth.
The afternoon (sun/wry) was still very hot.
The audience liked the presenters and their (wry/sun) sense of humour.
To enter the mansion we needed to find the long (key/cub) hidden under a stone.
The scared (cub/key) stayed close to the mother bear.
The sum was so hard she couldn't even (try/fad) to solve it.
The popular girl at school started a silly (fad/try) by wearing her socks inside-out.
The aggressive (cat/woo) refused to be touched by anyone.
The lobbyist will (woo/cat) politicians by giving them expensive gifts.
The cowboy used his loud (gun/aft) to scare the Indians.
The captain told the trainee to stand at the very (aft/gun) of the boat.
The man cheered when his favourite (son/coy) scored a goal.
The woman was very (coy/son) with the younger gentleman.
He stayed in the warm (sea/ark) until his fingers wrinkled.
He loaded up the wooden (ark/sea) with all of the animals.
She told a very (bad/doe) lie to her teacher.
He saw the graceful (doe/bad) leap through the field.
A few years (ago/bud) we traveled to Europe.
Watching the flowers (bud/ago) was magical.
She did not like having such a small (bed/tan) but she had no choice.
They lay in the garden to gently (tan/bed) themselves in the sun.
My birthday present came in a huge (box/din) that couldn’t even fit through my door.
Earl couldn't endure the awful (din/box) for too long.
It was during this (era/elf) that Christopher Columbus found the New World.
Cloze
Predictability
0.00
0.00
0.00
0.18
0.00
0.00
0.09
0.00
0.45
0.00
0.00
0.09
0.00
0.00
0.09
0.00
0.00
0.00
0.00
0.00
0.09
0.00
0.55
0.00
0.00
0.00
0.64
0.00
0.00
Violation
syntactic
syntactic
semantic
semantic
semantic
semantic
semantic
semantic
semantic
semantic
syntactic
semantic
semantic
semantic
syntactic
syntactic
syntactic
syntactic
semantic
Semantic
Syntactic
Semantic
Syntactic
Syntactic
Semantic
Semantic
Semantic
Semantic
Semantic
Target
word
class
closed
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
closed
open
open
open
open
open
open
Preview
word
class
open
closed
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
closed
open
open
open
open
open
48
Pair
66
67
67
68
68
69
69
70
70
71
71
72
72
73
73
74
74
75
75
76
76
77
77
78
78
79
79
80
80
Sentence
Number
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
Sentence
At the front of the line, a happy little (elf/era) was waiting to greet the children.
The officer asked if by any chance (she/sum) had seen anything.
He could tell nobody about his terrible (sin/she) for many years.
The man was told that he should (sit/toy) down and wait.
The child tried to steal the expensive (toy/sit) but he was caught.
In the forest, the giant (bee/tar) flew right past my head.
Eric got his shoes stuck in the thick (tar/bee) and had trouble getting them free.
For my sister's birthday she asked for an apple (pie/pro) instead of a cake.
They took lessons from the tanned (pro/pie) at the tennis club.
I told my friend (not/nub) to disturb my father while he was working.
She sharpened her pencil down to a tiny (nub/not) in the middle of class.
Arthur's sore (leg/pat) caused him a lot of pain.
He gave his colleague a friendly (pat/leg) on the back.
He asked to be immediately (let/mop) into the house, since it was raining so terribly.
My dad told me to use our heavy (mop/let) in order to clean up the spilled juice.
The man filled his small (bag/fog) with only the things he needed.
The bus driver was driving slow because heavy (fog/bag) limited his visibility.
They told him to sell (his/lap) car this year and use a bike instead.
He collapsed on the last (lap/his) of the race, just a few feet from the finish line.
I wish I could (fly/gin) so I could see the world from above.
The old man ordered (gin/fly) at the hotel bar.
The strange (guy/icy) in Anna's theatre class makes her feel uncomfortable.
The road was extremely (icy/guy) and the driver could not continue.
The specialist tried to see as many (ill/rip) people as possible.
It is easy to accidentally (rip/ill) this dress.
She really wanted to wear the blue (top/wit) that was on sale.
His biting (wit/top) won the comedian many admirers.
It was not time to leave (yet/cot) and so the kids just played in the street.
The child in the small (cot/yet) slept soundly.
Cloze
Predictability
0.09
0.00
0.00
0.00
0.00
0.00
0.00
0.36
0.00
0.00
0.09
0.00
0.36
0.00
0.00
0.27
0.00
0.27
0.09
0.00
0.00
0.00
0.00
0.00
0.00
0.09
0.00
0.45
0.00
Violation
Semantic
Syntactic
Syntactic
Syntactic
Syntactic
Semantic
Semantic
Semantic
Semantic
Syntactic
Syntactic
Semantic
Semantic
Syntactic
Syntactic
Semantic
Semantic
Syntactic
Syntactic
Syntactic
Semantic
Semantic
Syntactic
Syntactic
Syntactic
Semantic
Semantic
Syntactic
Syntactic
Target
word
class
open
closed
open
open
open
open
open
open
open
closed
open
open
open
open
open
open
open
closed
open
open
open
open
open
open
open
open
open
closed
open
Preview
word
class
open
open
closed
open
open
open
open
open
open
open
closed
open
open
open
open
open
open
open
closed
open
open
open
open
open
open
open
open
open
closed
49
Pair
81
81
82
82
83
83
84
84
85
85
86
86
87
87
88
88
89
89
90
90
91
91
92
92
93
93
94
94
95
Sentence
Number
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
Sentence
I will tell her I have (got/kit) the crisps and peanuts for the party.
When he bought the chemistry (kit/got) it didn’t come with instructions.
The school boy would (run/pox) everyday so he could stay in shape.
The expert was worried about the spread of the dangerous (pox/run) virus this year.
The boy felt he had become a real (man/sax) after his hard Summer work on the farm.
The magnificent (sax/man) solo was what jazz musician was famous for.
The purple (pen/hop) spilt all over my shirt.
I trained my pet so it will (hop/pen) on command.
I couldn't think of a good (end/ray) to the story, so I needed help.
She directed a strong (ray/end) of light onto the wall using a prism.
He broke the local (law/dye) and he's going to jail.
She could (dye/law) her hair if she wanted to look different.
They looked at the afternoon (sky/cut) and admired the clouds.
She accidentally (cut/sky) herself while making a collage.
The skyscraper didn’t look very (big/dab) from the airplane.
She added another (dab/big) of paint to her canvas, and then stopped for the day.
The strong (men/fun) lifted the wheels.
They knew that much (fun/men) was to be had at the carnival.
The class took a public (bus/err) to the museum.
Even the best mathematicians will sometimes (err/bus) in their calculations.
The bird watcher (saw/ore) the most amazing blue jay yesterday.
The explorers were very excited because they found (ore/saw) on the small island.
The spy crawled very (low/bun) to the ground so the guards couldn’t see him.
We only had a single (bun/low) and one tomato, so we couldn't make hamburgers.
He could (see/foe) the entire valley with his binoculars.
After his cunning (foe/see) had been defeated, the hero was relieved.
After spending a few minutes in the humid (air/lag) she collapsed.
Many countries still (lag/air) behind others in terms of environmental awareness.
We decided to stop (and/kid) help the stranded driver.
Cloze
Predictability
0.00
0.18
0.09
0.00
0.09
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.09
0.00
0.09
0.00
0.00
0.00
0.27
0.09
0.00
0.00
0.00
0.00
0.00
0.00
0.36
0.00
0.18
Violation
Syntactic
Syntactic
Semantic
Semantic
Semantic
Semantic
Semantic
Semantic
Semantic
Semantic
Semantic
Syntactic
Semantic
Syntactic
Syntactic
Semantic
Semantic
Syntactic
Syntactic
Semantic
Syntactic
Syntactic
Syntactic
Semantic
Syntactic
Semantic
Semantic
Semantic
Syntactic
Target
word
class
closed
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
closed
Preview
word
class
open
closed
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
50
Pair
95
96
96
97
97
98
98
99
99
100
100
101
101
102
102
103
103
104
104
105
105
106
106
107
107
108
108
109
109
Sentence
Number
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
Sentence
The old woman told the little (kid/and) to stop yelling.
At the desk, they will (ask/elm) her if she packed her own luggage.
She planted a tiny (elm/ask) tree near the local library.
The captain made the guard (arm/inn) himself for battle.
They decided to stay at a lovely (inn/arm) for the night.
In the military, it is customary to call one's superiors (sir/jog) as a sign of respect.
The woman was tired from her long (jog/sir) through the park.
During the flight Harriet (sat/dip) near the window so she could get a good view.
We went for a pleasant (dip/sat) after our long hot day.
He could not understand his brother's slightly (odd/par) behaviour these days.
They all tried to remain under (par/odd) but didn't succeed.
The team were very (sad/sew) because they lost the championship.
Agatha's grandmother will always (sew/sad) in her free time.
The student (ran/oak) so fast his shoes almost fell off.
From the top of the tall (oak/ran) the campers could see how big the forest truly was.
My friends took me to the fancy (pub/rev) for my birthday.
The teenager would (rev/pub) his engine at every stop-light.
The arena turned into (mud/jot) after the rain.
The student had to quickly (jot/mud) down the homework before the teacher erased it.
If the team stays disciplined they should at least (tie/tag) against their opponents.
The boy's neck itched because of the uncomfortable (tag/tie) on his shirt.
He found the shoes a tiny (bit/vow) too small.
You must (vow/bit) to be with another person for the rest of your life when you get married.
Before she could (say/tin) anything, the salesman cut her off.
They used (tin/say) to fix their toy boat.
The worker picked up his enormous (axe/pun) and cut down the large tree.
The speaker made a very clever (pun/axe) during dinner.
The fruit was very (red/tow) and looked delicious.
Tomorrow they will (tow/red) their caravan to the seaside.
Cloze
Predictability
0.00
0.09
0.00
0.00
0.09
0.45
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.82
0.00
0.00
0.00
0.27
0.09
0.00
0.00
Violation
Semantic
Syntactic
Syntactic
Syntactic
Semantic
Semantic
Semantic
Syntactic
Syntactic
Semantic
Syntactic
Syntactic
Syntactic
Syntactic
Syntactic
Semantic
Syntactic
Syntactic
Syntactic
Semantic
Semantic
Semantic
Syntactic
Semantic
Syntactic
Semantic
Semantic
Syntactic
Syntactic
Target
word
class
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
Preview
word
class
closed
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
open
51
Pair
110
110
111
111
112
112
113
113
114
114
115
115
116
116
117
117
118
118
119
119
120
120
Sentence
Number
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
Sentence
The teacher was very (mad/lob) when she found out that the students had cheated.
The instructor will simply (lob/mad) the ball at the weaker students.
To get hired for the competitive (job/zip) the woman had to work sixty hours a week.
The cars will (zip/job) past the observers during the race.
The girl asked her parents if she could play outside (too/hue) and they consented.
The curator pointed out the beautiful (hue/too) of the photo.
The child always asks very politely if he and his brother (may/zoo) go outside.
Our school went on a field trip to the renowned (zoo/may) last Friday.
There is now a strict (ban/wed) on smoking in restaurants.
The couple was finally (wed/ban) last autumn.
My friend misread the complicated (map/hug) so we got lost.
She gave a tight (hug/map) to her Mum when she saw her.
After my brother picked the wrong (way/sly) we got lost for hours.
The exceedingly (sly/way) thief had no problems stealing the painting.
Her friends took her to a small (bar/paw) for her birthday.
The injured (paw/bar) had to be cleaned and bandaged.
She wouldn't (fit/flu) into her old prom dress if she tried it on now.
The man thought he had severe (flu/fit) but he really just had a cough.
The child did not know the numbers after (ten/yew) yet but learned quickly.
There was a tall (yew/ten) tree behind the house.
The dogs were extremely (wet/wag) so we didn't let them in the house.
With a quick (wag/wet) of its tail the dog bounded in the house.
Cloze
Predictability
0.09
0.00
0.36
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.36
0.00
0.00
0.00
0.00
0.00
0.00
0.45
0.00
0.00
0.00
Violation
Syntactic
Syntactic
Semantic
Syntactic
Syntactic
Syntactic
Syntactic
Syntactic
Syntactic
Syntactic
Semantic
Semantic
Syntactic
Syntactic
Semantic
Semantic
Syntactic
Syntactic
Syntactic
Semantic
Syntactic
Syntactic
Target
word
class
open
open
open
open
closed
open
closed
open
open
open
open
open
open
open
open
open
open
open
closed
open
open
open
Preview
word
class
open
open
open
open
open
closed
open
closed
open
open
open
open
open
open
open
open
open
open
open
closed
open
open
52
Table A2: Word frequency and mean letter bigram (token) frequency for each target word (taken from the CELEX database using the
N-Watch software, Davis, 2005).
Sentence
Number
Pair
1
1
2
2
3
3
4
4
5
5
6
6
7
7
8
8
9
9
10
10
11
11
12
12
13
13
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
Target
Word
all
ant
buy
hag
aim
ape
boy
beg
but
mat
can
cow
day
dew
dry
duo
did
don
dog
dim
cup
nip
eye
emu
far
cap
Bigram
CELEX
token
frequency frequency
3597.49
3630.37
3.85
15047.34
126.2
2900.31
1.01
4274.34
40.89
1469.16
5.08
15.27
216.54
324.75
11.62
209.49
5412.79
7233.24
7.54
1406.73
2081.84
2908.21
22.51
1661.13
766.98
2573.04
5.2
836.78
89.55
256.79
0.67
57.24
1170.11
1321.28
79.22
245.05
71.73
134.3
15.7
1942.18
60.84
155.45
1.96
42.47
127.6
144.25
N/A
N/A
515.64
942.2
30.34
1295.7
53
Pair
14
14
15
15
16
16
17
17
18
18
19
19
20
20
21
21
22
22
23
23
24
24
25
25
26
26
27
27
Sentence
Number
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
Target
Word
for
rub
few
fin
gas
shy
get
gym
god
gag
had
oar
has
hem
her
hit
him
bug
hot
hen
how
hid
its
lip
lay
rum
met
pad
Bigram
CELEX
token
frequency frequency
8288.04
8394.62
12.74
160.43
585.08
1150.69
3.63
132.26
70.89
6587.67
18.04
2399.6
1169.94
1945.94
4.08
4.08
260.78
719.91
1.84
105.56
6255.03
7541.48
0.39
640.57
2123.13
10756.46
3.46
1947.06
3854.97
4054.32
91.34
4402.21
2515.31
5395.71
3.24
2837.49
139.55
3909.55
5.59
2431.92
1183.8
2286.02
11.96
4774.81
1552.12
1552.12
16.98
116.62
156.15
2351.93
5.53
167.97
148.16
1760.23
11.79
3442.71
54
Pair
28
28
29
29
30
30
31
31
32
32
33
33
34
34
35
35
36
36
37
37
38
38
39
39
40
40
41
41
Sentence
Number
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
Target
Word
new
nap
nor
net
now
hog
oil
ode
old
owl
one
rid
our
orb
pay
pea
per
nod
put
pew
own
sap
six
rot
war
spy
you
rug
Bigram
CELEX
token
frequency frequency
1062.07
1377.48
5.03
78.8
171.96
7779.9
32.35
1905.9
1802.12
5167.04
2.4
718.96
123.85
128.99
1.01
30.87
752.35
752.35
3.02
467.29
3455.7
3455.7
35.36
699.83
1286.54
2562.32
0.45
1.99
189.66
2300.55
1.68
333.85
363.97
2314.75
11.12
3706.87
687.26
4780.78
2.4
1034.94
916.03
923.8
2.57
847.23
211.96
394.11
8.32
3286.06
339.78
6841.37
8.38
9.24
7189.27
7190.19
11.68
163
55
Pair
42
42
43
43
44
44
45
45
46
46
47
47
48
48
49
49
50
50
51
51
52
52
53
53
54
54
55
55
Sentence
Number
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
Target
Word
win
soy
two
tug
why
wig
who
wad
was
tee
via
tub
use
web
tea
den
sat
tip
tax
urn
are
pup
act
lab
aid
awe
sun
wry
Bigram
CELEX
token
frequency frequency
62.35
122.96
2.18
227.01
1371.62
1371.62
5.42
35.6
620.28
1832.43
6.7
226.43
2395.59
2705.73
3.24
9519.41
10857.38
12734.38
4.08
779.91
19.61
21.17
7.88
33.03
485.98
485.98
5.81
42.49
88.77
295.11
9.05
494.19
228.04
1071.58
24.36
85.03
108.88
134.1
2.63
2.63
4424.36
4577.31
0.89
387.85
187.15
189.02
9.94
196.03
56.76
848.26
8.27
15.14
153.3
352.83
2.63
213.33
56
Pair
56
56
57
57
58
58
59
59
60
60
61
61
62
62
63
63
64
64
65
65
66
66
67
67
68
68
69
69
Sentence
Number
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
Target
Word
key
cub
try
fad
cat
woo
gun
aft
son
coy
sea
ark
bad
doe
ago
bud
bed
tan
box
din
era
elf
she
sin
sit
toy
bee
tar
Bigram
CELEX
token
frequency frequency
71.56
89.43
2.57
150.07
268.38
346.21
3.13
3620.25
41.28
1520.04
2.4
573.53
63.58
319.73
1.06
1.62
159.66
251.92
1.84
179.9
160.45
1045.89
2.85
2358.74
209.78
3510.2
2.74
98.94
225.14
354.47
3.85
2826.43
244.47
483.41
7.37
1761.53
78.66
236.15
4.47
733.32
23.24
31.84
0.34
4.33
4132.18
34687.18
25.31
349.21
119.94
564.58
14.58
842.51
6.59
781.89
4.64
707.48
57
Pair
70
70
71
71
72
72
73
73
74
74
75
75
76
76
77
77
78
78
79
79
80
80
81
81
82
82
83
83
Sentence
Number
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
Target
Word
pie
pro
not
nub
leg
pat
let
mop
bag
fog
his
lap
fly
gin
guy
icy
ill
rip
top
wit
yet
cot
got
kit
run
pox
man
sax
Bigram
CELEX
token
frequency frequency
12.57
138.43
5.81
10.26
5109.72
6790.56
0.73
32.49
63.52
348.64
19.05
430.54
393.3
1665.73
4.08
150.52
62.63
243.06
9.39
4214.7
5576.54
6894.42
18.66
257.58
50.95
56.79
15.25
90.03
56.48
157.94
9.11
35.2
62.63
1861.66
4.58
69.25
236.7
843.65
12.63
338.32
469.22
1982.37
10.73
3284.65
860.11
3803.6
8.49
319.69
229.89
387.61
1.51
74.1
1061.79
2794.89
N/A
832.31
58
Pair
84
84
85
85
86
86
87
87
88
88
89
89
90
90
91
91
92
92
93
93
94
94
95
95
96
96
97
97
Sentence
Number
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
Target
Word
pen
hop
end
ray
law
dye
sky
cut
big
dab
men
fun
bus
err
saw
ore
low
bun
see
foe
air
lag
and
kid
ask
elm
arm
inn
Bigram
CELEX
token
frequency frequency
19.44
692.34
5.08
805.94
496.7
14880.67
12.91
2263.97
166.65
488.64
5.7
83.3
77.09
80.22
177.88
4548.38
317.15
475.47
2.4
417.14
656.15
889.22
45.98
286.18
64.53
2842.29
1.4
18.47
387.88
1078.3
3.07
2221.76
144.58
1860.36
3.74
3062.1
1171.28
1530.69
13.91
4179.33
251.17
388.45
3.3
226.93
28767.93
29677.73
32.12
692.29
226.42
249.38
6.7
7.51
106.03
2410.33
9.44
14.3
59
Pair
98
98
99
99
100
100
101
101
102
102
103
103
104
104
105
105
106
106
107
107
108
108
109
109
110
110
111
111
Sentence
Number
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
Target
Word
sir
jog
sat
dip
odd
par
sad
sew
ran
oak
pub
rev
mud
jot
tie
tag
bit
vow
say
tin
axe
pun
red
tow
mad
lob
job
zip
Bigram
CELEX
token
frequency frequency
167.09
477.37
3.35
197.49
228.04
1071.58
8.66
690.7
59.72
102.29
5.25
764.66
46.2
4083.74
2.51
1746.97
111.23
1782.13
14.19
15.2
20.73
382.46
6.48
103.85
29.22
42.46
3.13
3388.07
35.47
154.66
5.7
118.38
240.67
587.37
5.2
1622.15
878.27
2941.59
28.66
127.65
5.64
5.64
1.34
609.64
188.94
415.51
3.41
2323.73
48.21
4418.89
1.45
395.7
244.25
299.89
2.35
40.67
60
Pair
112
112
113
113
114
114
115
115
116
116
117
117
118
118
119
119
120
120
Sentence
Number
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
Target
Word
too
hue
may
zoo
ban
wed
map
hug
way
sly
bar
paw
fit
flu
ten
yew
wet
wag
Bigram
CELEX
token
frequency frequency
1043.91
1239.14
4.58
85.19
1057.32
3276.73
8.72
538.75
13.8
1886.2
3.3
354.47
32.18
1182.38
4.64
53.44
1205.03
8377.26
5.7
31.99
67.88
832.16
3.18
437.27
69.94
347.62
4.36
29.84
226.48
653.6
1.56
1453.95
62.51
1397.65
1.34
6252.27